ChatGPT, Gemini, Siri and Perplexity answer questions by reading the web, resolving who each business is, and citing the sources they trust. Here is that pipeline — and exactly why each audit recommendation moves you into the answer.
AI assistants don't browse your site the way a person does. Their crawlers fetch raw pages, and their models read text, headings, links and machine-readable markup. Anything hidden behind heavy scripts, images of text, or blocked by robots.txt is effectively invisible.
Before a model can recommend you, it has to decide who you are: name, category, location, services, opening hours. It builds this 'entity' from your pages, your structured data (schema.org), and third-party sources. Conflicting or missing facts make the model uncertain — and uncertain models stay silent.
When someone asks 'best dentist near downtown open Saturday', the assistant retrieves candidate businesses and cites the sources it trusts. Pages with clear, quotable statements — prices, hours, service lists, reviews — are easy to cite. Vague marketing copy is not.
Finally the model picks who to mention. It favors entities it can describe confidently, with fresh, consistent, well-structured information. This is where visibility is won or lost — and where every gap found earlier in the chain compounds.
JSON-LD markup is the one format models can read with zero ambiguity. Adding LocalBusiness, Service, FAQPage or OpeningHoursSpecification markup turns 'probably a dentist' into 'verified dentist, open Saturdays'.
Typical fix: add schema.org/Dentist with openingHoursSpecification to location pages.
Assistants cite sentences, not slogans. A page that says 'Emergency appointments within 2 hours, £95 call-out fee' gives the model something it can repeat. 'We care about your smile' gives it nothing.
Typical fix: rewrite service pages around concrete facts — prices, times, areas served, guarantees.
If robots.txt blocks AI crawlers, or key pages error or redirect, the model works from stale or missing data. An llms.txt file gives assistants a curated map of your most important content.
Typical fix: allow GPTBot/ClaudeBot/PerplexityBot, publish llms.txt, fix broken service pages.
If your homepage says 'est. 2012' and your About page says 'est. 2014', or your hours differ between pages, the model lowers its confidence in everything you say — including the true parts.
Typical fix: reconcile names, dates, hours and service lists across every crawled page.
Most AI-visibility advice is generic. The [HUB] audit is the opposite: it reads your actual pages — up to 8 per site, plus robots.txt, llms.txt and your sitemap — and every recommendation must cite the page it came from.
Your hours appear as plain text on the contact page. Assistants can't confirm them, so 'open now' queries skip you.
No. SEO optimizes for a ranked list of links. AI assistants produce a single synthesized answer with a handful of mentions. The unit of competition is the citation, not the ranking — and the signals that win citations (structured data, quotable facts, entity consistency) only partially overlap with classic ranking factors.
Because every recommendation must be grounded in something real. Each finding cites the exact page and the exact line it is based on, and our system re-checks each quote against the live page text. If a quote can't be verified, it's flagged rather than trusted — so you never act on an invented claim.
It varies by assistant. Crawl-based engines (Perplexity, Copilot) can reflect changes in days to weeks. Models that rely on periodic retraining (base ChatGPT, Gemini) take longer, which is why repeated measurement matters: you see the change as it propagates instead of guessing.
Yes — the levers above are not secret. What the audit gives you is the diagnosis: which of the four stages is failing for your business, on which page, with the evidence attached. Most businesses we audit have 3–5 high-priority gaps, and they are rarely the ones the owner expected.
Yes. If your site is in German, we ask the assistants the customer questions in German, the way your customers ask. The report comes in the language you start the audit in: English here, German on the German site.
Run the free snapshot and get your first cited findings in about a minute.